Abstract 238: A Machine Learning Algorithm To Predict Refractory Ventricular Fibrillation

Abstract

Introduction: Guidelines for resuscitation of ventricular fibrillation (VF) out-of-hospital cardiac arrest (OHCA) advise drug therapy (epinephrine and antiarrhythmics) if a patient requires ≥3 shocks. Given that a higher number of shocks is associated with lower survival, the ability to predict which patients will require ≥3 shocks could enable preemptive, targeted intervention (antiarrhythmics or early hospital transport for invasive interventions) aimed at improving outcome. Aim: We sought to design an ECG-based algorithm to predict whether VF OHCA patients will require ≥3 shocks. Methods: We evaluated a cohort of adult VF OHCA from a metropolitan EMS system between 2008-2020. Patients were randomized 80%-20% for training-test. We trained an algorithm to predict whether a patient will require ≥3 shocks using 3-s ECG segments collected immediately before and within a minute after initial shock. The method uses a random forest classifier to predict shock count based on singular value decompositions of ECG wavelet transforms. Test performance was quantified by area under the receiver operating characteristic curve (AUC). Association between functional survival and shock count was evaluated with logistic regression. A sensitivity analysis assessed prediction of patients requiring ≥4 shocks. Results: Of 1376 VF OHCA patients, 311 (23%) were female; the median (IQR) shock count was 3 (1-6) and 591 (43%) achieved functional survival. Shock count was associated with lower functional survival with an odds ratio (95% CI) of 0.92 (0.89-0.95) for each additional shock. In 275 test patients, AUC (95% CI) was 0.80 (0.74-0.85) for predicting ≥3 shocks and 0.81 (0.75-0.85) for predicting ≥4 (Figure). Conclusions: A machine learning algorithm based on ECGs proximal to the initial shock predicts patients likely to experience refractory VF and high shock count, and could enable rescuers to preemptively administer interventions targeted to improve resuscitation outcome.